Group of diverse professionals collaborating in a modern office, discussing multi-agent communication strategies with digital

Editorial illustration for Study Finds No One-Size-Fits-All Strategy for Multi-Agent Communication

Study Finds No One-Size-Fits-All Strategy for...

Study Finds No One-Size-Fits-All Strategy for Multi-Agent Communication

By AI Daily Post Edited by Brian Petersen, Editor-in-Chief

June 6, 2026 • 2 min read

Why do multi‑agent systems built on large language models still choke on token bloat? Most designs rely on predefined roles, pipelines and turn‑taking schedules, yet they leave the actual messages between agents as free‑form text. The result? Rapid consumption of the shared context window, higher inference costs and a dip in overall efficiency.

The new study puts five typical messaging approaches to the test across two distinct MAS topologies. It discovers that a single, one‑size‑fits‑all rule doesn’t exist; performance swings with the nature of the task and the architecture.

Enter PACT—Protocolized Action‑state Communication and Transmission. Instead of dumping raw output into the conversation log, PACT re‑encodes each agent’s reply as a concise state update before it becomes part of the shared history. The authors report that this modest compression yields comparable—or even better—task results while slashing token counts.

In practice, OpenHands sees a 10 % drop in tokens per resolved item, and SWE‑agent’s resolve rate stays flat even as input tokens are cut in half. The code is openly released for anyone to explore.

We analyze five common inter-agent communication strategies across two MAS topologies, finding that no fixed strategy is universally optimal. Instead, effective inter-agent messages consistently preserve action-centered information needed by downstream agents. Building on this, we propose the PACT (Protocolized Action-state Communication and Transmission), which treats inter-agent communication as a public state-update problem and projects each raw agent output into a compact action-state record before it enters shared history.

Across different MAS topologies, PACT consistently improves the performance-cost trade-off, achieving comparable or stronger task performance with substantially fewer tokens. The gains extend to production coding harnesses: PACT lifts OpenHands' resolve rate at -10% tokens-per-resolved, and is resolve-neutral on SWE-agent while halving input tokens.

What Should Agents Say? Action-state Communication for Efficient Multi-Agent Systems - ArXiv AI (cs.AI)

Why this matters

We’ve seen that letting agents chat freely can blow up token counts, eat the shared context window, and push inference costs higher. Token bloat, real cost. The study’s five‑strategy sweep across two MAS topologies shows no single recipe works everywhere; instead, messages that keep the focus on the next action tend to help downstream agents.

That insight nudges us toward more disciplined prompting rather than abandoning structure altogether. The authors’ PACT framework builds on this observation, but the paper leaves open how PACT scales to larger fleets or more complex tasks. For developers, the takeaway is modest: prioritize action‑oriented content when designing inter‑agent protocols, and monitor token budgets closely.

Founders should be wary of assuming a one‑size‑fits‑all communication layer will save compute. Researchers might explore whether the identified pattern holds beyond the tested topologies. Unclear whether PACT will become a standard component or remain a niche solution; further validation will be needed before we adopt it broadly.

Study Finds No One-Size-Fits-All Strategy for...

Further Reading

Latest News